Flux Stylizer V2.1 (Important Update)
5.0
1 reviews
Description
ComfyUI Workflow: Transforming Photos into Studio Ghibli-Style Characters
Notes: Go to the change log for information about the new version.
I recommend Using the workflow with ComfyUI Desktop, as there are discrepancies between the Desktop and Portable versions, especially in the custom nodes.
___
IMPORTANT: If you have already created custom styles, before downloading any updates, I recommend that you backUps the JSON strings.
___
1. Introduction
Recently, OpenAI introduced an exciting new feature in ChatGPT: the ability to generate images that closely adhere to prompts and modify uploaded photos. As a result, many users have been experimenting with transforming real images into Studio Ghibli-style illustrations. However, ChatGPT’s free tier limits users to only three image modifications per day.
To overcome this limitation and provide unlimited image transformations, I created a workflow for ComfyUI that replicates the same effect, allowing for greater flexibility and continuous image generation.
Interfaccia Precedente
___
2. Workflow Overview
The workflow follows these key steps:
- Image Input: The user provides a photo as input.
- Auto-Prompting with Florence2: The Florence2 model analyzes the input image and generates an automatic description.
- Prompt Refinement: The generated description is then modified to include key elements of Studio Ghibli’s art style and the trigger word required for the LoRA model.
- Image Generation with Flux Model: The core of this workflow is the Flux1.dev Hyper 8 steps GGUF Q8_0 model, which processes the refined prompt to generate a Studio Ghibli-style illustration.
- Preprocessing with ControlNet Depth: To ensure the output remains faithful to the input photo, ControlNet Depth is applied using the Flux_ControlNet-Union-Pro model as a preprocessor.
- Processing with Custom Advanced Sampler: The workflow utilizes the Custom Advanced Sampler for optimized image rendering.
- Postprocessing: To enhance the final output, the workflow applies: 15% monochromatic Gaussian noise for a vintage effect. 50% saturation boost to make colors more vivid.
___
3. Models Used
Below is a list of the primary models used in this workflow, along with their corresponding links:
- Flux1.dev Hyper 8 steps GGUF Q8_0 The core model used for image generation. Flux1.dev Hyper 8 on HuggingFace
- Florence2 (Auto-Prompting) Generates a description of the input image. In case you do not already have The Florence2 model, it will be downloaded automatically from the custom node.
- Studio Ghibli LoRA A LoRA fine-tuned to replicate the distinctive Studio Ghibli art style. Download LoRA on Civitai
- Flux_ControlNet-Union-Pro (ControlNet Depth Preprocessor) Ensures structural consistency between the input and output images. Flux_ControlNet-Union-Pro on HuggingFace
- Custom Advanced Sampler Handles image rendering for high-quality output.
___
___
4. Conclusion
This workflow allows users to transform real photos into Studio Ghibli-style illustrations while maintaining key facial and structural details. By integrating the models listed above, the system not only delivers high-quality results but also removes the generation restrictions found in the free version of ChatGPT.
This is the perfect solution for artists and AI enthusiasts who want to experiment freely and generate unlimited images.
___
5. Change Log
09/04/2025 Important Update
IMPORTANT: If you have already created custom styles, before downloading any updates, I recommend that you backUps the JSON strings.
Flux Stylizer – v2.1
🔧 Major Enhancements
• JSON-based Custom Style Management
A new system has been implemented to manage custom styles through structured JSON strings. This includes:
- A simplified syntax, easier to write and maintain.
- An internal parser that auto-formats entries to ensure valid JSON.
- A predefined structure that improves both readability and integration with the workflow UI.
• Persistent Style Storage via File System
Custom style definitions are now saved in external files, allowing persistence across sessions. Key points:
- A new subfolder models/JSON/ must be manually created to store the JSON configuration file.
- On first startup, the workflow will auto-generate a data file if none exists.
- The workflow now loads styles from the file, not from the in-workflow JSON node, although the node is still present as a safety fallback.
⚙️ Workflow Behavior Options
New "Mode" Node for Customization
The workflow now includes a centralized "Mode" control node with several toggles:
- AutoPrompt: Automatically generates prompts from input images using the Florence2 template. Can be disabled to allow manual prompt input.
- Creative Mode: Prioritizes stylistic freedom over pose accuracy. When disabled, a depth map is used to closely match the input image’s structure.
- Ultimate SD Upscale: Enables 2x upscaling via the "Ultimate SD Upscale" node. Increases quality, but also processing time.
- Save Image / Save Compare: Control saving of generated images and comparison images. Self-explanatory toggle switches.
📝 Improved Documentation
- The manual now includes a complete tutorial on how to add new styles using the new JSON format.
- Examples and detailed breakdowns are provided to help users write valid custom style entries.
- Instructions cover advanced behavior such as handling apostrophes in strings using escape characters (\').
🐛 Bug Fixes
• LoRA Path Resolution
Fixed an issue where LoRAs placed in the root of the models/loras/ directory were not recognized by the LoRA Switcher node.
→ LoRA models now work whether they are in subfolders or directly in the root path.
07/04/2025 Bug fix
Bug Fix Update – LoRA Path Handling
I've just uploaded a small update to the workflow to fix an issue that occurred when LoRA models were placed directly in the root of the models/loras/ folder. Previously, the LoRA Switcher node couldn't correctly detect styles unless they were inside a subfolder. This behavior has now been corrected—LoRAs placed directly in the root directory should work as expected.
Thanks again to everyone who provided feedback and helped identify the problem. Let me know if you encounter any further issues!
03/04/2025 Version 2.0
In this workflow update, I’ve worked on making it more customizable and configurable for the end user, allowing the addition of new styles easily and quickly. Now, the workflow is no longer limited to creating the Studio Ghibli style, it can generate any style you want.
As a result, I’ve also updated the name: it could no longer remain the original one, so it is now called Flux Stylizer V2.0.
To enable this flexibility, I structured the data in JSON format, using nodes that read and extract values associated with specific keys. Additionally, I’ve integrated a manual directly into the workflow, accessible via a "NOTE" node, which explains in detail how to:
- Add new styles
- Insert LoRA models and their trigger words
- Use specific strings to reinforce styles in prompts
If you have any questions about adding new styles, feel free to ask in the comments, and I’ll reply as soon as possible.
Another key update is the ability to write a custom prompt instead of using Florence2’s automatic one. I’ve also introduced a method based on IPAdapter to preserve image consistency and color accuracy. Moreover, I’ve added an option to activate a creative mode. All these details can be found in another "NOTE" node, located on the right side of the workflow interface.
I hope this update is well received, as it makes the workflow more versatile, allowing you to replicate any style you desire.
Below are 3 trials with different styles, autoprompt ON and Creative Mode ON
------
------
Discussion
(No comments yet)
Loading...
Reviews
No reviews yet
Versions (5)
- latest (6 months ago)
- v20250407-174325
- v20250407-172336
- v20250403-151446
- v20250331-130609
Node Details
Primitive Nodes (108)
> Saturation (1)
Anything Everywhere3 (1)
Bjornulf_LoadTextFromPath (1)
Bjornulf_ShowStringText (2)
Bjornulf_ShowText (3)
Compare-🔬 (1)
DownloadAndLoadFlorence2Model (1)
DualCLIPLoaderGGUF (1)
EmptySD3LatentImage (1)
Fast Groups Bypasser (rgthree) (3)
Florence2Run (1)
FluxGuidance (1)
GetNode (29)
If ANY return A else B-🔬 (2)
Int-🔬 (3)
JSON extractor [Crystools] (2)
Label (rgthree) (9)
LoadJsonFromText (7)
MarkdownNote (3)
ModelSamplingFlux (1)
PrimitiveString (5)
ReduxAdvanced (1)
Reroute (6)
SetNode (20)
UnetLoaderGGUF (1)
VaeGGUF (1)
easy isFileExist (1)
Custom Nodes (68)
- ImageNoiseGaussian (1)
- Lora (2)
- CR Text Concatenate (7)
- CR Simple Image Compare (1)
ComfyUI
- KSamplerSelect (1)
- RandomNoise (1)
- BasicScheduler (1)
- BasicGuider (1)
- SamplerCustomAdvanced (1)
- StyleModelLoader (1)
- CLIPVisionLoader (1)
- VAEDecode (1)
- ControlNetLoader (1)
- PreviewImage (5)
- LoraLoader (2)
- SaveImage (3)
- LoadImage (1)
- ControlNetApplyAdvanced (1)
- UpscaleModelLoader (1)
- LayerUtility: PurgeVRAM (2)
- AIO_Preprocessor (1)
ComfyUI_tinyterraNodes
- ttN text (4)
- > Scale Image to Side (1)
- > Int to Text (1)
- GetTextFromJson (5)
- Big Int Switch [Dream] (1)
- Big Text Switch [Dream] (1)
- GetImageSizeAndCount (1)
- String Replace (mtb) (1)
- ShowText|pysssss (4)
- Replace Text _O (6)
- UltimateSDUpscale (1)
- Seed Everywhere (1)
- Text to Conditioning (2)
- Save Text File (1)
- Boolean To Text (1)
- Text Dictionary To Text (1)
Model Details
Checkpoints (0)
LoRAs (4)
3DMM_V12.safetensors
3DMM_V12.safetensors ACE_Plus\comfyui_local_lora16.safetensors ACE_Plus\comfyui_portrait_lora64.safetensors ACE_Plus\comfyui_subject_lora16.safetensors Al_Rio_Style-10.safetensors Android_Style.safetensors Android_Style_V1.1.safetensors BHDA_Style.safetensors BHStyle_V21_MC-000005.safetensors BHStyle_V21_MC-000006.safetensors BHStyle_V21_MC-000007.safetensors BHStyle_V22_MC.safetensors BHStyle_V2_MC.safetensors Brown_Style\Brown_Style-000001.safetensors Brown_Style\Brown_Style-000002.safetensors Brown_Style\Brown_Style-000003.safetensors Brown_Style\Brown_Style-000004.safetensors Brown_Style\Brown_Style-000005.safetensors Brown_Style\Brown_Style-000006.safetensors Brown_Style\Brown_Style-000007.safetensors Brown_Style\Brown_Style-000008.safetensors Brown_Style\Brown_Style-000009.safetensors Brown_Style\Brown_Style.safetensors Concept_controls\SDXL\age.pt Concept_controls\SDXL\cartoon_style.pt Concept_controls\SDXL\chubby.pt Concept_controls\SDXL\clay_style.pt Concept_controls\SDXL\cluttered_room.pt Concept_controls\SDXL\curlyhair.pt Concept_controls\SDXL\dark_weather.pt Concept_controls\SDXL\eyebrow.pt Concept_controls\SDXL\eyesize.pt Concept_controls\SDXL\festive.pt Concept_controls\SDXL\fix_hands.pt Concept_controls\SDXL\long_hair.pt Concept_controls\SDXL\muscular.pt Concept_controls\SDXL\pixar_style.pt Concept_controls\SDXL\professional.pt Concept_controls\SDXL\repair_slider.pt Concept_controls\SDXL\sculpture_style.pt Concept_controls\SDXL\smiling.pt Concept_controls\SDXL\stylegan_latent1.pt Concept_controls\SDXL\stylegan_latent2.pt Concept_controls\SDXL\suprised_look.pt Concept_controls\SDXL\tropical_weather.pt Concept_controls\SDXL\winter_weather.pt CreateLogo.safetensors CyberWorld_v1.1.safetensors DatBush.safetensors De_Witte_Style.safetensors DirtDraw.safetensors ExplosionMagic-20.safetensors Final_Fantasy_Logo_style.safetensors Flux.1\Akira_Toriyama_Style.safetensors Flux.1\Alex_Ross_StyleF1.safetensors Flux.1\Anime_style.safetensors Flux.1\Art style Impressionist.safetensors Flux.1\BerserkLoraFlux.safetensors Flux.1\Cybernetic_Style_-_World_Morph_-_Flux.1.safetensors Flux.1\DisneyStyleSMALLERV2.safetensors Flux.1\Elden_Ring_-_Yoshitaka_Amano.safetensors Flux.1\Eldritch_Comics_for_Flux_1.1.safetensors Flux.1\Eldritch_Parchment_Art_for_Flux_1.2.1.safetensors Flux.1\Eldritch_Pastels_for_Flux_1.0.2.safetensors Flux.1\FLUX.1-Turbo-Alpha.safetensors Flux.1\Flux Turbo 8 steps.safetensors Flux.1\Flux_Fantasy_Portraits_V3.safetensors Flux.1\Grzegorz_Rosiński_Style.safetensors Flux.1\Hyper-FLUX.1-dev-8steps-lora.safetensors Flux.1\JimLee_Flux1-000007.safetensors Flux.1\Loisel Style.safetensors Flux.1\Mezzotint_Artstyle_for_Flux_-_by_Ethanar.safetensors Flux.1\Persona5-Guy90-Flux-V1.safetensors Flux.1\PetkaV3.safetensors Flux.1\RPGv6-itr15.safetensors Flux.1\Studio_Ghibli_Flux.safetensors Flux.1\baishi_style_flux.safetensors Flux.1\chargen-celshading-lora_v2_dev.safetensors Flux.1\dirtysketch.safetensors Flux.1\drawdewitte.safetensors Flux.1\fantasy-flux-v1.safetensors Flux.1\flux1-depth-dev-lora.safetensors Flux.1\flux_realism_lora.safetensors Flux.1\hcklknst_lora.safetensors Flux.1\ligne_claire_f1d.safetensors Flux.1\messysketch-000005.safetensors Flux.1\shadman_flux.safetensors Flux.1\watercolor_flux_v1.1_rank_16_bf16 .safetensors GameIconResearch_chest2_Lora.safetensors GoodHands-beta2.safetensors Graphic_Novel_Illustration-000007.safetensors Hyper SD\Hyper-SDXL-1step-lora.safetensors Hyper SD\Hyper-SDXL-2steps-lora.safetensors Hyper SD\Hyper-SDXL-4steps-lora.safetensors Hyper SD\Hyper-SDXL-8steps-CFG-lora.safetensors Hyper SD\Hyper-SDXL-8steps-lora.safetensors Immoral_City_v1.safetensors Insectoid_LORA.safetensors KitchenStyle.safetensors LowRA dark theme.safetensors LowRA.safetensors Russo Alessandro.safetensors SD1.5\Hyper\Hyper-SD15-1step-lora.safetensors SD1.5\SDXLrender_v2.0.safetensors SD1.5\Steampunk.safetensors SD1.5\animatediff\v3_sd15_adapter.ckpt SD1.5\epiCRealismHelper.safetensors SD1.5\lcm-lora-sdv1-5.safetensors SD1.5\lcm\pytorch_lora_weights.safetensors SDXL Lightning\sdxl_lightning_2step_lora.safetensors SDXL Lightning\sdxl_lightning_8step_lora.safetensors SDXL Offset Lora\sd_xl_offset_example-lora_1.0.safetensors SDXL-Caricaturized-Lora.safetensors SDXL\1980s_Fantasy_Style_SDXL.safetensors SDXL\CLAYMATE_V2.03_.safetensors SDXL\ClassipeintXL1.9.safetensors SDXL\ClassipeintXL2.1.safetensors SDXL\DetailedEyes_V3-XL.safetensors SDXL\EnvyScifiStreamlineXL01.safetensors SDXL\FaeTastic2.safetensors SDXL\Fantasy_Races_XL.safetensors SDXL\HandFineTuning_XL.safetensors SDXL\Harrlogos_v2.0.safetensors SDXL\Jean-Baptiste Monge Style.safetensors SDXL\JuggerCineXL2.safetensors SDXL\Professional_Portrait.safetensors SDXL\SDXL Detail.safetensors SDXL\SDXLFaeTastic2400.safetensors SDXL\Stained Glass Portrait.safetensors SDXL\TCD-SDXL-LoRA.safetensors SDXL\add-detail-xl.safetensors SDXL\alb3nizrodrigu3zv1PonySDXL.safetensors SDXL\bl3uprint.safetensors SDXL\detailed_hands-000002.safetensors SDXL\diffusers_xl_recolor_testXL-fp16.safetensors SDXL\figurine.safetensors SDXL\fractal-art.v1.safetensors SDXL\hand 4.safetensors SDXL\lcm-lora-sdxl.safetensors SDXL\lego_v2.0_XL_32.safetensors SDXL\rppc-final.safetensors SDXL\sd_xl_turbo_lora_v1-64dim.safetensors SDXL\silhouette style v2.safetensors SDXL\ttplanetSDXLControlnet_v20Fp16.safetensors SDXL\xl_more_art-full_v1.safetensors Scifi\CircuitryTechXL-000008.safetensors Scifi\Cyber_Tech_sdxl.safetensors Scifi\DAYU_box.ckpt Scifi\DaVinciTech-25.safetensors Scifi\HackedTech-20.safetensors Scifi\JJsScifiStyle_XL.safetensors Scifi\LEDarrayTech-20.safetensors Scifi\NeoFuturisticTechXL.safetensors Scifi\OrokinTech-20.safetensors Scifi\OverwatchTech-28.safetensors Scifi\PrecursorTech-20.safetensors Scifi\PurityTech-30.safetensors Scifi\UltraBlackhole_Tech_XL.safetensors Scifi\WireTechXL.safetensors Space Engineers\DHCAS v1.0.safetensors Spaceship_art.safetensors ThunderMagic-22.safetensors Type3Tech-21.safetensors Xris_Hawkins_Cartoon_Anime-000007.safetensors add_detail.safetensors akiraToriyamaStyle_toriyamaAkira3.safetensors anime3d.safetensors aodai_SD_chiasedamme_v02.safetensors artgerm_v2.1.safetensors baoping_lora_v1.1-000009.safetensors bettergunAK47WIP_betaWIP.safetensors biomechanicals_v2.safetensors blindbox_V1Mix.safetensors de_witte_style_ultimate_28200.pt de_witte_v2_4200.pt de_witte_v2_4200_txt.pt detailed_eye-10.safetensors epiNoiseoffset_v2.pt epiNoiseoffset_v2.safetensors epiNoiseoffset_v2Pynoise.safetensors film_grain.safetensors funnyCreatures_v1.safetensors game Icon Institute_anfg_v3-000018.safetensors goodhands_Beta_Gtonero.safetensors hKHK416Lora_v20.safetensors ipadapter\ip-adapter-faceid-plus_sd15_lora.safetensors ipadapter\ip-adapter-faceid-plusv2_sd15_lora.safetensors ipadapter\ip-adapter-faceid-plusv2_sdxl_lora.safetensors ipadapter\ip-adapter-faceid_sd15_lora.safetensors ipadapter\ip-adapter-faceid_sdxl_lora.safetensors jimLeeDCComicsMarvel_offset.safetensors jim_lee_offset_right_filesize.safetensors koreanDollLikeness.safetensors lksecaiV2.safetensors logoarchive_v10.safetensors maplestoryStyle_v30.safetensors marbleshLora_marblesh.safetensors more_details.safetensors openjourneyLora.safetensors paintedMiniature_v35.safetensors polyhdron_all_in_one_eyes_hands_skin_fin.safetensors polyhedron_all_eyes.safetensors polyhedron_all_sdxl-000004.safetensors polyhedron_men_eyes.safetensors polyhedron_skinny_all.pt sp4c3sh1p.safetensors spybgsToolkit_spybgtkCEnh.safetensors stanley_lau_artgerm.safetensors steampunkai10MBLora_10mb.safetensors theovercomer8sContrastFix_sd15.safetensors theovercomer8sContrastFix_sd21768.safetensors to8contrast-1-5.safetensors warhammer40000AddonFor_wh40k.safetensors wowifierXL (art by mooncryptowow).safetensors 不是青花瓷第三代-但是机甲.safetensors
Flux.1\Alex_Ross_StyleF1.safetensors